국내 최대 기계 및 로봇 연구정보

기술보고서

기술보고서 게시판 내용
타이틀	Parallel computations of incompressible viscous flow in a lid-driven square cavity and program performance on the NWT computer system.
저자	Shigeki Hatayama
Keyword	square cavity problem, non-staggered grid, consistent finite-difference approximation, iterative method/Jacobi/red-black/CG/ADI, SIMD/MIMD computing, synchronization/data-transfer overhead, program performance, maximum performance, half-performance grain size,
URL	http://send.nal.go.jp/send/jpn/dlpdf.php3/naltr0001363t.pdf?id=NALTR0001363T
보고서번호	NAL TR-1363T
발행년도	1998.10
출처	NAL (National Aerospace Laboratory of Japan)
ABSTRACT	The NWT computer system available at the NAL since February 1993 comprises two system administrators, n processing elements (where n was 140 at the beginning, and is 166 at present) and a crossbar network, and operates as a distributed-memory message-passing MIMD computer. Each processing element itself is a vector computer. This paper reports parallel computations of incompressible viscous flow in a lid-driven square cavity on the NWT computer system. In order to obtain numerical solutions of this flow, consistent finite-difference approximations on non-staggered grids and four iterative solution methods are used. Computations are performed on the Reynolds number range of Re = 0 ~105, and effects of the Reynolds number, number of processing elements (pe) in the parallel processing, solution method and grid size on the computational results are examined. Actual rates of the parallelised square cavity programs on the NWT computer system are measured, and two characteristic parameters of these programs are estimated for the cases that the actual rate is considered as a function of pe and that the actual rate is considered as a function of the grid size. Measurements of the maximum actual rate and estimations of the speedup and efficiency against pe on the NWT computer system are indicated as well.